The problem with the noise discussion, is that context is needed. Most people don't relate a dB number to something realistic, and even those charts are horribly misleading (ie. it might say it is at whispering level or talking in a room or something, but my BM is way more quiet than a whisper).
And... in my world, MacBook Pro fans are intolerably loud (and annoying!). It also depends so much on environment. In our current home, we've got a fairly busy street/road outside. With the windows open, I'm only going to hear fans if they are fairly noisy or certain pitches. (Those little fans in laptops, unfortunately, cut right through a lot of other noise.)
With the windows closed, I can hear a lot more, but even then, enough of that background noise gets in, that I wouldn't be able to do any great testing. Hopefully we'll move to a quieter place in the not too distant future.
But, in our last place (when I got the BM and did most of the testing), it was quite quiet. When the BM was cranking 100%, there was a very slight 'air movement' type sound you could hear if you knew what you were listening for.
(This description by John Siracusa of his Mac Pro is relevant, as it is a lot like he describes, only I'd never be able to hear it with headphones on... so quieter than a Mac Pro.
https://overcast.fm/+R7DUKc3N4/54:55 )
Do you ever get the little fans to spin up on the GPU? I'd imagine they'd have to be somewhat similar to the MBP noise. The reason the BM is so darn quiet, is there is a huge fan spinning at very low RMPs, working with natural heat-rise in a vertical tunnel. I just don't think there is any other practical way to pull that kind of silence off.
But, that's good to hear you're happy with the noise level. I hope I get to try one some day and see for myself. Otherwise, it's just to hard to base it off descriptions, I guess. I've heard people say they can barely hear MBP fans... and I suppose if their environment is noisy, or maybe they are comparing it to a PS4 'taking off' or a gaming PC, maybe that is true in context.
Yes, though in the real-world (vs theoretical), it doesn't end up being that bad. I think it was either AppleInsider or Barefeats that did some extensive testing, and it was something like 8% loss worst-case, or something pretty low like that (don't quote me... I'd have to find the article again). Where it was bad (though still quite usable), was if you're trying to run the eGPU and then display on the MBP display. The data has to go out and return, so much more bottle-necked. So, unless it is some kind of portability thing, you'd absolutely want to be running any kind of eGPU as:
computer --> eGPU --> display off eGPU port.